First a dataset of the relevant Alphaproteobacteria (“alphas” for short) was constructed by downloading all the alphaproteobacterial genomes available in NCBI genome database. All animal obligate pathogens were excluded, since they show extensive genome reductions, linked with intracellular lifestyle [4,19], as well as the SAR11 clade due to the extensive gene loss described for this group [1]. A total of 92 genomes were then analyzed (Figure 1 and Supplementary Material S1), and divided into three groups: (i) solely free-living, (ii) plant-associated and (iii) symbiont (that is a sub-set of plant-associated) combining the information available on GOLD database [20,21], Bergey's manual of systematic bacteriology [22] and bibliographic search on Pubmed. Plant-associated bacteria include 27 genomes (2 pathogens, 7 associated and 18 symbionts), all but two (25/27) grouped within the order Rhizobiales (Figure 1), the only exceptions are the species Gluconacetobacter diazotrophicus and Azospirillum B510 which fall in the order Rhodospirillales; of course we cannot exclude that among the 65 nonplant-associated bacteria some could have also experienced the plant environment, even if those putative events have not been reported.

A quick look at these genomes shows a wide range of genome sizes, spanning from 2.9 Mbp (Parvularcula bermudensis HTCC2503) to 9.2 Mbp (the magnetotactic bacterium Magnetospirillum magnetotacticum MS1). The average genome size (±standard deviation) of the dataset is 5.04 ± 1.53 Mbp. Plant-associated bacteria have significant (P < 0.0001, one-way ANOVA) larger genomes (6.73 ± 1.26 Mbp) than nonplant-associated ones (4.34 ± 0.99 Mbp), as previously noticed [19]. The same trend was observed considering only the order Rhizobiales, which accounts for near half of the entire dataset (42 out of 92 genomes with average length of 5.83 ± 1.53 Mbp), with plant-associated Rhizobiales having genomes larger than those of nonplant-associated Rhizobiales (6.81 Mbp and 4.47 Mbp, respectively, P < 0.0001). Average GC content is 63.1 ± 4.4% and is similar for plant-associated and nonplant-associated genomes (63.03% and 62.71%, respectively, P < 0.8) and ranges from 45.2% (Hirschia baltica ATCC49814) to 71.1% (Phenylobacterium zucineum HLK1); within the Rhizobiales we observed the same trend: an average of 63.1% with 63.0% for plant-associated and 63.1% for nonplant-associated).

To answer this question we first peformed a genome clusterization of all protein coding genes present in the 92 genomes, obtaining 40,960 groups of orthologs (out of a total number of 434,411 proteins analyzed). Next, starting from these groups of orthologs, we proceeded trying to extract four groups named as: (i) Alpha Core (common to all the analyzed organisms), (ii) Plant-Associated (common to and exclusive of plant associated bacteria), (iii) Plant-Symbionts (common to and exclusive of plant symbionts), and (iv) NonPlant-Associated (common to and exclusive of nonplant-associated bacteria). Since genetic elements inside bacteria are prone to horizontal gene transfer and therefore genes that may be specific for a certain life-style may be found also in other bacterial species, by chance or because they might carry out a different function, we developed an orthologs-species clustering approach capable of taking into account this dynamical behavior. This “Fuzzy orthologs-species clusterization” analysis (see Materials and Methods) sorted out 998 orthologous groups for the Alpha Core subset, while life-style specific subset of NonPlant-Associated, Plant-Associated and Plant-Symbionts accounted for 88, 15 and 73 orthologous groups, respectively (Figure 2 and Supplementary Material S2). As expected, the Non-Plant subset of orthologous groups was found to be inconsistent, since a series of random subsets having the same number of species (see materials and methods) showed a similar number of orthologous groups. This finding is not surprising, since the NonPlant subset consists of species with no unique distinctive habitat, varying from soil to marine and freshwater organisms. On the contrary, for the other subsets, the number of orthologous groups generated from random species list was zero; the same result was observed when sampling only inside the order Rhizobiales (where most of the Plant-associated species are), retrieving 1.7 orthologous groups on average (data not shown). Notably, when we did not apply fuzziness to the orthologs-species clusterization, we did not find anyorthologous groups in any subset, but only in the Alpha Coreone. This discrepancy suggests that the plant association behavior may not be dependent on a strongly conserved defined set of genes strictly common to all the species of the subset. However, interestingly, by applying the “Fuzzy orthologs-species clusterization”, the larger Plant-Associated subset was found to contain only 15 orthologous groups, while in contrast the smaller Plant Symbionts subset contains 73 orthologous groups. This finding again confirms that the generic plant association behavior does not require a large repertoire of specific genes, while the symbiotic interaction in α-rhizobia is more dependent on the presence of specific gene traits [23].

An overview of the distribution among the COGs (Cluster of Orthologous Groups) categories of the genes present in the Alpha Core, Plant-Associated and Plant Symbionts subsets is reported in Figure 3. Regarding the Plant-Associated and the Plant Symbionts subsets, COG categories related to basic cell functions (L, Replication, recombination and repair; B, Chromatin structure and dynamics D, Cell cycle control, cell division, chromosome partitioning; V, Defense mechanisms) are not represented, since they are mostly present in the Alpha Core subset, as expected; on the other hand, COG categories poorly or not characterized (S and X), are the most represented in both plant related subsets. The categories related to regulation of gene expression (K and J) and energy production and conversion (C) show a slightly higher proportion in the Plant-Associated subset. Regarding the Plant Symbionts subset, a slightly higher over-representation of the carbohydrate transport and metabolism category (G) was observed, suggesting the key role of carbohydrate metabolism for establishing nitrogen fixing symbiosis (i.e., for the formation of the so-called Nod factors as well as for bacteroid trophism [23]). Indeed, some plant symbionts, as for instance Sinorhizobium meliloti, contain large genomic regions or replicons mainly devoted to carbohydrate transport and metabolism [24-26]. However, the percentages of COG categories represented in the different subsets are not statistically different (Spearman Rank Correlation and Chi-square test with Monte Carlo simulation, data not shown).

The analysis of the GO (Gene Ontology) categories in the Plant-Associated subset (Supplementary Material S3) shows that the most represented biological process is related to electron carrier activity (4 groups), while for the Plant Symbionts the functions encoded appear to be more heterogeneous (Figure 4). In particular, proteins involved in many process were found, ranging from symbiosis specific functions, like nodulation (4 orthologous groups: 3672, 4012 and 4954 encoding NodA, NodC, and the NodJ protein respectively, plus group 3700 also encoding NfeD another protein necessary for nodulation [27]) to trascriptional regulation and to more general biological functions (especially oxidation-reduction), with a slightly higher presence of transport-related functions (11 groups, with 3 of them probably involved in osmolarity control). As reported in Figure 4, sugar transport (4 groups out of 11 transporter) and metabolism (5 groups), are highly represented in the symbiont subset. Within this category, of particular interest is group 2572 encoding for MocE, a Rieske non-heme iron oxygenase essential in the catabolism of rhizopines (3-O-methylscyllo-inosamine, 3-O-MSI) a nodule-specific compounds that confer an intraspecies competitive nodulation advantage to strains able to utilize them [28]. Another intriguing group is 3933 which encodes for a protein belonging to the senescence marker protein 30 (SMP-30)/gluconolaconase superfamily which contains many mammalian sequences [29]; this protein was found to accommodate multiple functions [30], among which calcium regulation (as a regucalcin) [31]; that is particularly intriguing as the involvement of Ca²⁺ in the symbiotic signaling pathway activated by flavonoids was found in Rhizobium leguminosarum bv. viciae [32]. Another interesting function which could be related to the plant symbiosis is cell motility, encoded by orthologous group 4140 (protein FliG). No nitrogen-fixation related proteins were found as exclusive, due to the presence in alphas of free-living nitrogen-fixing species (Xanthobacter autotrophicus [22]) and to the presence of the fix signaling module also in Caulobacter [33].

Interestingly, none of the functions previously putatively associated with endophytic life-style was detected, as type IV pili [34] or other metabolic or hormonal-related activities [35]. This is possibly due to the wide range of associations of our Plant-Associated subset, which includes also symbiotic and rhizospheric interaction; absence of type IV pili could be also linked to their involvement in many other processes in other species not engaged in plant interactions.

Once the list of life-style related orthologous groups was defined, we looked for their presence in the other branches of the bacterial taxonomic tree, in order to understand if such functions are specific for alphas or are widespread in other taxas, thus giving an insight into the evolutionary pathways of those functions. Each life-style associated orthologous group, was then used as a query on the GenBank database to find homologous sequences in all bacterial taxa; results of the analysis are reported in Figure 5 and Supplementary Material S4. Figure 5 offers an overview of the proportion of the orthologous groups occurring in each subset which have hits in the different bacterial classes. As expected, most of the hits were scored within Proteobacteria, in particular in the classes Beta and Gamma, possibly reflecting both a higher phylogenetic proximity and a general bias of the database which is abundant in sequences from members of such classes (Alphaproteobacteria were excluded from the analysis).

A high proportion of Alpha Core genes have at least one hit in almost all the taxa probed by the analysis, with an average of 10.5% of the Alpha Core subset having an hit in the selected taxa, while on average, only 4.2% and 5.0% of the Plant Associated and Plant Symbionts orthologous groups have at least a hit in each taxa; this observation suggests that plant association genes tend to be conserved only inside Alphaproteobacteria and to a lesser extent inside Beta- and Gamma-Proteobacteria (36% average), while just few genes have an homolog in phylogenetically distant species, where they might not be related to a plant association behavior.

To further elucidate this point, all the taxa found by this approach were investigated for their plant-association life-style, according to the GOLD database annotation; 33.3% of the Plant-Associated orthologous groups have at least one hit in species associated with plants, followed by the Plant Symbionts (30.1%) and the Alpha Core (24.6%). The two plant related subsets have plant association hits only inside the Protebacteria class (Beta- and Gamma-Proteobacteria), while the Alpha Core hits are distributed in a broader range, including Actinobacteria, Cyanobacteria and Firmicutes (Supplementary Material S5); the results of this analysis then imply that the plant association related genes are rather specific of the Proteobacteria class, while the housekeeping genes exhibit an higher degree of sequence conservation across all the bacterial phylogenetic tree.

To shed some light on the possibility that Plant-associated specific genes are involved in plant association at a broader taxonomic level, the protein coding genes found as exclusively present in Plant-Associated alphas were checked for their presence in the genomes of four known and fully-sequenced plant-associated Proteobacteria, in particular in the class of Beta-Proteobacteria the strains Cuprividus taiwanensis [17] and Azoarcus sp. BH72 [34] and in the class of Gamma-Proteobacteria the species Enterobacter sp. 638 [35] and Klebsiella pneumoniae 342 [36]. Results of the comparison are shown in Table 1. Interestingly, two out of 15 orthologous groups are present in all the four strains selected, namely orthologous group 2149 (Transcriptional regulator) and 2774 (endoribonuclease l-psp), suggesting that regulation (either by transcriptional regulation and RNA stability) may play pivotal roles in establishing the association with plant. Moreover 5 other genes were found to be present in at least 2 of the 4 species investigated. A previous work dealing with the description of the genome sequence of the β-rhizobium Cuprividus taiwanensis [17], found no gene both common and specific to all rhizobia, suggesting that symbiotic association with plants evolved with multiple strategies, even though genes preferentially associated (on a statistical basis) with plant symbiosis were detected. However, our findings suggest that these two genes, putatively needed in Alphaproteobacteria to establish a successful interaction with plants, are also present in the Beta and Gamma-Proteobacteria model organisms for plant association. Endoribonuclease L-PSP is involved in single-stranded mRNA cleavage in Leishmania infantum, a parasite that in its life cycle alternates two stages, and is hypothesized to be involved in specific post-transcriptional regulation of gene expression [37]; characterization of mutants for those orthologs is however necessary to fully elucidate their role in plant association in such a broad taxonomic background.

To construct our reference phylogenetic tree, all 16S rRNA gene sequences were aligned using MUSCLE [38], alignment was manually checked. The alignment was used with the software Mega 5.05 [39] to generate a phylogenetic tree. A Model test (Supplementary Material S6) was performed before running the Maximum Likelihood algorithm, with 1,000 bootstrap replicates and the Tamura-Nei model of evolution.

The 92 genomes were clustered together using the approach proposed by Kim and collaborators [40], using the PanGenomer software (available upon request); a total number of 8,464 pairwise InParanoid analyses with no thresholds were generated and the results were merged in a single file as an input for MCL [41], using an inflation factor of 5.0, a pruning threshold of 30,000 and a selection number of 5,000. To test the clusterization the presence of 5 well known orthologous genes, involved in cell cycle regulation and DNA replication (ctrA, dnaA, rpoE, gyrB, dnaQ), was used as a positive control (referred to as group 4, 315, 158, 524 and 26 respectively).

The obtained orthologous groups were mapped to the four species subsets (Figure 1) looking at the species from which each protein belonging to that group came from, using a so-called “Fuzzy” approach: an orthologous group was regarded as specific for one of the four subsets when its species list was comprised between 80% and 110% of the subset list. The biological value of each subset was tested generating 10 random organism lists with the same length of the subset and looking at how many orthologous groups were then retrieved.

The orthologous groups belonging to the four subsets were annotated, using ten proteins from each group (selected randomly) to speed-up the analysis. Each protein was mapped to the COG database [42] using rpsblast 2.2.25+ and an e-value threshold of 1e-10; the domain content and the GO [43] annotation were obtained using Iprscan 4.8 [44] with the InterPro database release 33.0.

The protein sequence similarity across the bacterial kingdom of each orthologous group was inspected using TaxonomyBlaster (available upon request); the same proteins used for the annotation were analyzed using a series of taxonomically-restricted portions of the NCBI nr database (downloaded on 1 July 2011): all the taxonomic classes (excluding the “environmental samples”) inside the Bacteria kingdom were iteratively used; the Proteobacteria class was further divided into the distinct classes, excluding the Alphaproteobacteria and the Proteobacterial “environmental samples”. BLAST was run using the BLOSUM45 matrix, the soft masking option, a fixed database size of 500,000,000 and the Smith-Waterman local optimal alignments option. Those hits showing an e-value below 1e-10, a query coverage above 66% and an homology index above 0.33, were retained. The obtained species were marked as Plant-Associated looking at the available information in the GOLD database [20].